定位尺度和像元空间关系对GF-1亚像元定位精度影响分析

亚像元定位技术对地表遥感信息提取及农业遥感定量化发展具有重要意义。针对当前国内外亚像元定位研究多集中于亚像元定位模型而缺少模型定位精度影响因素分析的现状,该文开展了定位尺度因素(如重建尺度、影像空间分辨率)和像元空间关系等对农业区域多光谱遥感影像亚像元定位模型精度影响的定量分析。以中国吉林省白城地区洮南市和内蒙古自治区兴安盟突泉县交界农业区为研究区域,以典型的空间引力模型为核心模型,以具有相同光谱分辨率的高分一号(GF-1)卫星8、16 m空间分辨率多光谱遥感影像为基础数据,对重建尺度、影像空间分辨率和像元空间关系等因素对遥感亚像元定位精度的影响进行了探讨。结果表明,对于8 m空间分辨率GF-1遥感影像,当重建尺度为5时,在邻接空间关系下的亚像元定位可达到最佳效果,即由40 m空间分辨率遥感影像重建8 m空间分辨率遥感影像的总体精度为74.67%,Kappa系数为0.604;对于16 m空间分辨率GF-1遥感影像,当重建尺度为4时在象限空间关系下的亚像元定位可达到最佳效果,即由64 m空间分辨率重建16 m空间分辨率遥感影像的总体精度为74.65%,Kappa系数为0.623。此外,重建尺度、影像空间分辨率和像元空间关系对亚像元定位精度具有波动影响,3个因素对应的亚像元定位总体精度最大变幅分别为18.08%、4.39%和0.08%,对应Kappa系数变化最大幅度分别为0.268、0.049和0.006。因此,在不同精度影响因素下,基于空间引力模型的GF-1亚像元定位精度影响因素轻重等级依次为重建尺度影像空间分辨率像元空间关系,这可为遥感亚像元定位模型选取、模型参数设置以及适宜的遥感数据选择提供一定参考。

Influence of mapping scales and pixel spatial relationships on GF-1 sub-pixel mapping accuracy

Abstract: Sub-pixel mapping technology is significant to land cover information extraction and development of quantitatively agricultural remote sensing. Sub-pixel mapping accuracy is affected not only by sub-pixel mapping model, but also by many other factors such as reconstruction scale, pixels spatial relationship and image spatial resolution. These factors increase complexity and uncertainty of sub-pixel mapping results. At present, most relevant researches mainly focus on sub-pixel mapping model itself but research on influencing factors of sub-pixel mapping accuracy is ignored. Therefore, in this paper, analysis on main influencing factors of multispectral remote sensing image sub-pixel mapping accuracy was carried out. The study region located at the junction area of Taonan County of Baicheng City in Jilin Province and Tuquan County of Xing'an League in Inner Mongolia Autonomous Region. Among them, Spatial Gravity Model (SGM) was applied in the sub-pixel mapping experiments which was one of the mainstream model in present sub-pixel mapping research field. Both 8 m and 16 m spatial resolution GF-1 multispectral remote sensing images were used as the data sources which were in the same sensor and had the same spectral resolution. Effects of reconstruction scale, pixels spatial relationship (touching, quadrant, surrounding) and image spatial resolution on the sub-pixel mapping accuracy in experiments were studied thoroughly. In the experiment of 8 m spatial resolution GF-1 remote sensing image, when sub-pixels was in touching spatial relationship and degradation scale was five, the experiment could achieve the best result of sub-pixel mapping. The overall accuracy and Kappa coefficient of 8 m spatial resolution remote sensing image reconstruction from 40 m spatial resolution remote sensing image were 74.67% and 0.604, respectively. In the experiment of 16 m spatial resolution GF-1 remote sensing image, when sub-pixel was in quadrant spatial relationship and degradation scale was four, the experiment could achieve the best result of sub-pixel mapping. The overall accuracy and Kappa coefficient of 16 m spatial resolution remote sensing image reconstruction from 64 m spatial resolution remote sensing image were 74.65% and 0.623, respectively. By analyzing sub-pixel mapping accuracy range caused by reconstruction scale, pixels spatial relationship and spatial resolution, the overall accuracy maximum range of these three factors were 18.08%, 4.39% and 0.08%, respectively. The corresponding Kappa coefficient maximum range was 0.268, 0.049 and 0.006, respectively. Therefore, by analyzing the sub-pixel mapping model mechanism, the order for affecting sub-pixels mapping accuracy based on SGM model was reconstruction scale > spatial resolution> pixels spatial relationship. The experiment results of this paper could provide a reference for choosing sub-pixel mapping models, setting model parameters and selecting the appropriate remote sensing data to obtain better sub-pixel mapping accuracy.